<![CDATA[Student dropout negatively impacts both university performance and the futures of individual students. Identifying at-risk students during the early stages of their studies is crucial for effective academic intervention. This study aims to develop an Early Warning System (EWS) using machine learning to predict dropout potential based on academic and demographic data from the first three semesters. To address the extreme class imbalance in the dataset, a cost-sensitive learning approach using class weighting was applied. Three classification algorithms were evaluated: Random Forest, XGBoost, and Logistic Regression. Performance evaluation using the Area Under the Precision-Recall Curve (AUC-PR) metric demonstrated that Logistic Regression outperformed the tree-based ensemble models, achieving the highest score of 0.6454. Feature interpretation revealed that low parental income, poor minimum attendance, and early course failure were the primary risk factors for dropping out. Furthermore, the best model was successfully implemented into a web-based Decision Support System (DSS) prototype dashboard. This system provides university management and academic advisors with an interactive tool for conducting proactive, data-driven early interventions.]]>
Copyrights © 2026
